tau2-bench

Sierra · MIT · license verified · open · 461 MB

Agent, tool, and user benchmark with retail, airline, and telecom domains. The repository is about 461 MB. This site does not mirror the repository. domains.json is our own short summary of those three domains, under MIT, with attribution to Sierra Research.

A trained model is a possible later option. It would be hosted on Hugging Face, not on this site.

URL

https://github.com/sierra-research/tau2-bench

License URL

https://github.com/sierra-research/tau2-bench/blob/main/LICENSE

Format

Git repository

Relevance

agent tool-use

Mirror allowed

yes

Data period

not stated

Verified

2026-10-20

Sample

/datasets/tau2-bench/domains.json

All datasets. card.json. README.md.